!eWOErHSaiddIbsUNsJ:nixos.org

NixOS CUDA

290 Members
CUDA packages maintenance and support in nixpkgs | https://github.com/orgs/NixOS/projects/27/ | https://nixos.org/manual/nixpkgs/unstable/#cuda57 Servers

Load older messages


SenderMessageTime
17 Jun 2024
@ss:someonex.netSomeoneSerge (back on matrix)I'd suggest get an MWE based on that and also reach out with runpod's support asking why they won't mount a driver compatible with the host's kernel14:16:36
@ss:someonex.netSomeoneSerge (back on matrix)(this conversation has happened here before: neither putting drivers into an image nor mounting the host's drivers is "correct": the driver in the image might not be compatible with the kernel running on the host, and the driver from the host might not be compatible e.g. with the libc in the image, et cetera)14:18:14
@grw00:matrix.orggrw00

great thanks, will try this and get back to you. it's

            Cmd = [ "${inputs.nix-gl-host.defaultPackage.x86_64-linux}/bin/nixglhost" "${my-bin}/bin/executor" ];

?

14:37:50
@grw00:matrix.orggrw00i guess i need to build some matrix of images with compat versions and choose which one based on cuda/kernel version in instance metadata14:39:17
@ss:someonex.netSomeoneSerge (back on matrix)
In reply to @grw00:matrix.org
i guess i need to build some matrix of images with compat versions and choose which one based on cuda/kernel version in instance metadata
You can build a single image with NixGL
14:39:55
@ss:someonex.netSomeoneSerge (back on matrix)Note: NixGL and nixglhost are different tools 🙃14:40:03
@ss:someonex.netSomeoneSerge (back on matrix) * You can build a single image with NixGL (and multiple drivers) 14:40:37
@grw00:matrix.orggrw00ah 😓14:41:32
19 Jun 2024
@hexa:lossy.networkhexapython312 default migration has starrted14:04:34
@ss:someonex.netSomeoneSerge (back on matrix)stupid new faiss not building with cuda =\14:19:56
21 Jun 2024
@search-sense:matrix.orgsearch-sense

Hello, NixOS community, I want to install python311Packages.tensorrt

TensorRT> command, and try building this derivation again.
TensorRT> $ nix-store --add-fixed sha256 TensorRT-8.6.1.6.Linux.x86_64-gnu.cuda-11.8.tar.gz
TensorRT> ***
error: builder for '/nix/store/140c5c8lpa30r3jrxxbw74631831prrw-TensorRT-8.6.1.6.Linux.x86_64-gnu.cuda-11.8.tar.gz.drv' failed with exit code 1;

but the cuda is 12.2 on my system, is it compatible?

> nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2023 NVIDIA Corporation
Built on Tue_Aug_15_22:02:13_PDT_2023
Cuda compilation tools, release 12.2, V12.2.140
Build cuda_12.2.r12.2/compiler.33191640_0
04:53:46
@search-sense:matrix.orgsearch-sense

Is anyone interested to add latest tensorrt-10.1.0 to NixOS ?

searching for dependencies of /nix/store/gknr686xg6ggafkdfy5323bc7f1m5yf7-tensorrt-10.1.0.27-lib/lib/stubs/libnvinfer_vc_plugin.so
    libstdc++.so.6 -> found: /nix/store/bn7pnigb0f8874m6riiw6dngsmdyic1g-gcc-13.3.0-lib/lib
    libgcc_s.so.1 -> found: /nix/store/pd8xxiyn2xi21fgg9qm7r0qghsk8715k-gcc-13.3.0-libgcc/lib
setting RPATH to: /nix/store/bn7pnigb0f8874m6riiw6dngsmdyic1g-gcc-13.3.0-lib/lib:/nix/store/pd8xxiyn2xi21fgg9qm7r0qghsk8715k-gcc-13.3.0-libgcc/lib:$ORIGIN
auto-patchelf: 1 dependencies could not be satisfied
error: auto-patchelf could not satisfy dependency libcudart.so.12 wanted by /nix/store/799sv915xqi5b8n14hdkbbp6h06rrjz7-tensorrt-10.1.0.27-bin/bin/trtexec
auto-patchelf failed to find all the required dependencies.
Add the missing dependencies to --libs or use `--ignore-missing="foo.so.1 bar.so etc.so"`.
error: builder for '/nix/store/7rqkwg91vnk5d3p4vaym0z0pskkmj4r8-tensorrt-10.1.0.27.drv' failed with exit code 1;
       last 10 log lines:
       >     libgcc_s.so.1 -> found: /nix/store/pd8xxiyn2xi21fgg9qm7r0qghsk8715k-gcc-13.3.0-libgcc/lib
       > setting RPATH to: /nix/store/bn7pnigb0f8874m6riiw6dngsmdyic1g-gcc-13.3.0-lib/lib:/nix/store/pd8xxiyn2xi21fgg9qm7r0qghsk8715k-gcc-13.3.0-libgcc/lib:$ORIGIN
       > searching for dependencies of /nix/store/gknr686xg6ggafkdfy5323bc7f1m5yf7-tensorrt-10.1.0.27-lib/lib/stubs/libnvinfer_vc_plugin.so
       >     libstdc++.so.6 -> found: /nix/store/bn7pnigb0f8874m6riiw6dngsmdyic1g-gcc-13.3.0-lib/lib
       >     libgcc_s.so.1 -> found: /nix/store/pd8xxiyn2xi21fgg9qm7r0qghsk8715k-gcc-13.3.0-libgcc/lib
       > setting RPATH to: /nix/store/bn7pnigb0f8874m6riiw6dngsmdyic1g-gcc-13.3.0-lib/lib:/nix/store/pd8xxiyn2xi21fgg9qm7r0qghsk8715k-gcc-13.3.0-libgcc/lib:$ORIGIN
       > auto-patchelf: 1 dependencies could not be satisfied
       > error: auto-patchelf could not satisfy dependency libcudart.so.12 wanted by /nix/store/799sv915xqi5b8n14hdkbbp6h06rrjz7-tensorrt-10.1.0.27-bin/bin/trtexec
       > auto-patchelf failed to find all the required dependencies.
       > Add the missing dependencies to --libs or use `--ignore-missing="foo.so.1 bar.so etc.so"`.
       For full logs, run 'nix log /nix/store/7rqkwg91vnk5d3p4vaym0z0pskkmj4r8-tensorrt-10.1.0.27.drv'.
07:59:22
@search-sense:matrix.orgsearch-sense

export NIXPKGS_ALLOW_UNFREE=1 && nix-build -A cudaPackages.tensorrt

       > setting RPATH to: /nix/store/bn7pnigb0f8874m6riiw6dngsmdyic1g-gcc-13.3.0-lib/lib:/nix/store/pd8xxiyn2xi21fgg9qm7r0qghsk8715k-gcc-13.3.0-libgcc/lib:$ORIGIN
       > auto-patchelf: 1 dependencies could not be satisfied
       > error: auto-patchelf could not satisfy dependency libcudart.so.12 wanted by /nix/store/799sv915xqi5b8n14hdkbbp6h06rrjz7-tensorrt-10.1.0.27-bin/bin/trtexec
       > auto-patchelf failed to find all the required dependencies.
       > Add the missing dependencies to --libs or use `--ignore-missing="foo.so.1 bar.so etc.so"`.

11:03:14
@ss:someonex.netSomeoneSerge (back on matrix)
In reply to @search-sense:matrix.org

Hello, NixOS community, I want to install python311Packages.tensorrt

TensorRT> command, and try building this derivation again.
TensorRT> $ nix-store --add-fixed sha256 TensorRT-8.6.1.6.Linux.x86_64-gnu.cuda-11.8.tar.gz
TensorRT> ***
error: builder for '/nix/store/140c5c8lpa30r3jrxxbw74631831prrw-TensorRT-8.6.1.6.Linux.x86_64-gnu.cuda-11.8.tar.gz.drv' failed with exit code 1;

but the cuda is 12.2 on my system, is it compatible?

> nvcc --version
nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2023 NVIDIA Corporation
Built on Tue_Aug_15_22:02:13_PDT_2023
Cuda compilation tools, release 12.2, V12.2.140
Build cuda_12.2.r12.2/compiler.33191640_0
You can use cudaPackages.overrideScope to plug in the trt release compatible with your cuda, but also I think trt was originally introduced in Nixpkgs with a logic to select the compatible release in each cuda package set automatically. Evidently, that must be have broken
15:51:43
@ss:someonex.netSomeoneSerge (back on matrix)
In reply to @ss:someonex.net
Nvidia prevents unattended downloads, of course it broke
...primarily because of^^^ and because no one seems to be actively using Nixpkgs' in-tree trt expression?
15:52:47
@lcw:matrix.orgLucas joined the room.17:13:01
@lcw:matrix.orgLucas

Does anyone have nsight_systems working?

I am using CUDA to develop progrmans on NixOS 24.05 and it is working great. Now I want to profile my code.
Using the following flake

``` I was able to
get `ncu` working
19:46:16
@lcw:matrix.orgLucas *

Does anyone have nsight_systems working?

I am using CUDA to develop progrmans on NixOS 24.05 and it is working great. Now I want to profile my code.
Using the following flake

{
  description = "nsight_systems";

  inputs = {
    # nixpkgs.url = "github:NixOS/nixpkgs/release-24.05";
    # nixpkgs.url = "github:NixOS/nixpkgs/nixpkgs-unstable";
    nixpkgs.url = "github:ConnorBaker/nixpkgs/feat/cudaPackages-fixed-output-derivations";
  };
  outputs = { self, nixpkgs }:
    let
      system = "x86_64-linux";
      pkgs = import nixpkgs { system = system; config.allowUnfree = true; };
    in
    {
      devShells.${system}.default = pkgs.mkShell {
        nativeBuildInputs = [
          pkgs.cudaPackages.nsight_systems
          pkgs.cudaPackages.nsight_compute
        ];
      };
    };
}

```
I was able to get `ncu` working.
19:47:05
@lcw:matrix.orgLucas *

Does anyone have nsight_systems working?

I am using CUDA to develop progrmans on NixOS 24.05 and it is working great. Now I want to profile my code.
Using the following flake

{
  description = "nsight_systems";

  inputs = {
    # nixpkgs.url = "github:NixOS/nixpkgs/release-24.05";
    # nixpkgs.url = "github:NixOS/nixpkgs/nixpkgs-unstable";
    nixpkgs.url = "github:ConnorBaker/nixpkgs/feat/cudaPackages-fixed-output-derivations";
  };
  outputs = { self, nixpkgs }:
    let
      system = "x86_64-linux";
      pkgs = import nixpkgs { system = system; config.allowUnfree = true; };
    in
    {
      devShells.${system}.default = pkgs.mkShell {
        nativeBuildInputs = [
          pkgs.cudaPackages.nsight_systems
          pkgs.cudaPackages.nsight_compute
        ];
      };
    };
}

I was able to get ncu working.

19:47:39
@lcw:matrix.orgLucas

But when I try to run nsys-ui I get a dialogue box with the error message

Failed to load plugin: QuadDPlugin

Cannot load library /nix/store/hzp2wmqbqihx4slp353ixs405ry6li4f-cuda12.5-nsight_systems-2024.2.3.38-bin/nsight-systems/2024.2.3/host-linux-x64/Plugins/QuadDPlugin/libQuadDPlugin.so: /nix/store/hzp2wmqbqihx4slp353ixs405ry6li4f-cuda12.5-nsight_systems-2024.2.3.38-bin/nsight-systems/2024.2.3/host-linux-x64/Plugins/QuadDPlugin/libQuadDPlugin.so: undefined symbol: _ZNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEED1Ev, version Qt_6

Some functionality will be disabled
19:50:21
@lcw:matrix.orgLucas I do have to launch ncu and ncu-ui via their full paths /nix/store/0v8ydp1hq7ynncwir4hv5hkpna629iw0-cuda12.5-nsight_compute-2024.2.0.16/nsight-compute/2024.2.0/ncu and /nix/store/0v8ydp1hq7ynncwir4hv5hkpna629iw0-cuda12.5-nsight_compute-2024.2.0.16/nsight-compute/2024.2.0/ncu-ui, respectively. 19:52:26
@lcw:matrix.orgLucas *

Does anyone have nsight_systems working?

I am using CUDA to develop programs on NixOS 24.05 and it is working great. Now I want to profile my code.
Using the following flake

{
  description = "nsight_systems";

  inputs = {
    # nixpkgs.url = "github:NixOS/nixpkgs/release-24.05";
    # nixpkgs.url = "github:NixOS/nixpkgs/nixpkgs-unstable";
    nixpkgs.url = "github:ConnorBaker/nixpkgs/feat/cudaPackages-fixed-output-derivations";
  };
  outputs = { self, nixpkgs }:
    let
      system = "x86_64-linux";
      pkgs = import nixpkgs { system = system; config.allowUnfree = true; };
    in
    {
      devShells.${system}.default = pkgs.mkShell {
        nativeBuildInputs = [
          pkgs.cudaPackages.nsight_systems
          pkgs.cudaPackages.nsight_compute
        ];
      };
    };
}

I was able to get ncu working.

20:02:11
@glepage:matrix.orgGaétan Lepage

connor (he/him) (UTC-5) could you please give a shot at

nix build github:GaetanLepage/nixpkgs/jaxlib#python311Packages.jaxlib
20:48:30
@glepage:matrix.orgGaétan Lepage *

connor (he/him) (UTC-5) could you please give a shot at the following ?

nix build github:GaetanLepage/nixpkgs/jaxlib#python311Packages.jaxlib
20:48:48
@glepage:matrix.orgGaétan Lepage *

connor (he/him) (UTC-5) could you please give a shot at the following ?

nix build github:GaetanLepage/nixpkgs/jaxlib#python311Packages.jaxlibWithCuda
20:48:50
@aidalgol:matrix.orgaidalgol
In reply to @ss:someonex.net
...primarily because of^^^ and because no one seems to be actively using Nixpkgs' in-tree trt expression?
Sorry, I have not been using TensorRT in a while, so I'm not catching these, even though I put it in nixpkgs to begin with. :S
21:19:55
@lcw:matrix.orgLucas
In reply to @lcw:matrix.org

Does anyone have nsight_systems working?

I am using CUDA to develop programs on NixOS 24.05 and it is working great. Now I want to profile my code.
Using the following flake

{
  description = "nsight_systems";

  inputs = {
    # nixpkgs.url = "github:NixOS/nixpkgs/release-24.05";
    # nixpkgs.url = "github:NixOS/nixpkgs/nixpkgs-unstable";
    nixpkgs.url = "github:ConnorBaker/nixpkgs/feat/cudaPackages-fixed-output-derivations";
  };
  outputs = { self, nixpkgs }:
    let
      system = "x86_64-linux";
      pkgs = import nixpkgs { system = system; config.allowUnfree = true; };
    in
    {
      devShells.${system}.default = pkgs.mkShell {
        nativeBuildInputs = [
          pkgs.cudaPackages.nsight_systems
          pkgs.cudaPackages.nsight_compute
        ];
      };
    };
}

I was able to get ncu working.

I was able to get nsight_systems working from nixpkgs.url = "github:mcwitt/nixpkgs/fix/nsight_systems";.
21:26:13
@ss:someonex.netSomeoneSerge (back on matrix)
In reply to @lcw:matrix.org

But when I try to run nsys-ui I get a dialogue box with the error message

Failed to load plugin: QuadDPlugin

Cannot load library /nix/store/hzp2wmqbqihx4slp353ixs405ry6li4f-cuda12.5-nsight_systems-2024.2.3.38-bin/nsight-systems/2024.2.3/host-linux-x64/Plugins/QuadDPlugin/libQuadDPlugin.so: /nix/store/hzp2wmqbqihx4slp353ixs405ry6li4f-cuda12.5-nsight_systems-2024.2.3.38-bin/nsight-systems/2024.2.3/host-linux-x64/Plugins/QuadDPlugin/libQuadDPlugin.so: undefined symbol: _ZNSt7__cxx1112basic_stringIcSt11char_traitsIcESaIcEED1Ev, version Qt_6

Some functionality will be disabled
Looks likw a lwftover vendored qt library that we oughtta relink
21:45:46
@lcw:matrix.orgLucas
In reply to @ss:someonex.net
Looks likw a lwftover vendored qt library that we oughtta relink
Oh cool. Is there an example of relinking the libraries that I can follow? Is it this https://github.com/ConnorBaker/nixpkgs/blob/9ee229fe705580b62fc9011f5d8cc78e87f85971/pkgs/development/cuda-modules/overrides/cuda/nsight_systems.nix#L102-L121 ?
22:19:56
22 Jun 2024
@search-sense:matrix.orgsearch-sense
In reply to @ss:someonex.net
...primarily because of^^^ and because no one seems to be actively using Nixpkgs' in-tree trt expression?

the essence of the problem is this:

       > error: auto-patchelf could not satisfy dependency libcudart.so.12 wanted by /nix/store/799sv915xqi5b8n14hdkbbp6h06rrjz7-tensorrt-10.1.0.27-bin/bin/trtexec
       > auto-patchelf failed to find all the required dependencies.```
16:12:11

Show newer messages


Back to Room ListRoom Version: 9